A different model for each script.
About 989K parameters per word model. Choose a language and only its model and dictionary load—not all five.
SMALL BY DESIGN. LOCAL BY DEFAULT.
One keyboard.
Five ways to belong.
Your language, in its own script. Type with familiar English letters; tiny on-device models do the rest.
5Indian languages
7.72Mparameters, all add-ons
0inference-server calls
YOU TYPE ROMAN LETTERS
YOUR SCRIPT BENGALI
RUNNING ON YOUR DEVICE
Loading only the selected language…What you type stays yours.
Models run in a JavaScript Web Worker. After assets load, typing needs no network and no inference server.
Suggestions, not certainty.
These are transliterators, not translators or chatbots. Ambiguous spellings and names can fail. Hindi alone has an optional sentence LM.
Attribution required. Non-commercial.
Any use — research, demos, derivatives, or redistribution — must credit Sourojit Dhua (sourojitdotcom) and link to the model page. This bundle is CC-BY-NC-4.0, which requires attribution and forbids commercial use. Bengali, Telugu, Kannada and Tamil training includes Samanantar-derived Aksharantar rows; the upstream Samanantar card states CC-BY-NC-4.0, so this release does not certify those weights as commercially clean. The original Hindi word and Hindi sentence components remain separately available under Apache-2.0.
Ready-to-copy credit line: AksharaLP-MultiIndic-8M by Sourojit Dhua (sourojitdotcom) — https://huggingface.co/sourojitdotcom/AksharaLP-MultiIndic-8M